1. Field of Invention
The invention presented herein provides methods for analyzing images.
2. Description of Related Art
The outstanding characteristics of modern light detectors have made it possible to retrieve high quality images from a wide variety of platforms. However, their high-fidelity nature does not help in disentangling the images of objects blended together in crowded fields, those that are partially obscured or barely visible in the noise of low signal-to-noise (SNR) data, or adversarial efforts to purposefully deceive image detectors. Although certain indicators of quality are readily apparent, such as pixilated objects that are unresolved, obvious blurriness, or uncertainty in visual characteristics due to low SNR, obtaining a realistic understanding of the limitations of image data is often extremely difficult. Failing to detect an object does not provided enough data to be certain that the object is not actually present.
One approach to analyzing crowded fields in astronomy is artificial star analysis. In this approach, a light pattern appropriate for a star of a given brightness, color and location in the sky is added to the digitized image data of a real star field. The modified data re then analyzed in the usual way, and the parameters derived from the artificial star are compared to the known input parameters. This process is then repeated thousands of times for stars with randomly chosen characteristics. The deviations between the output of the analysis program and the known characteristics of the artificial stars are then used to evaluate the relation between the results for the real stars and the true underlying stellar populations in the field.
There are several limitations inherent in this method. The physics of target/background interaction in terrestrial images is more complex than artificial star insertion into an astronomical image because terrestrial objects are typically more complex than stellar profiles and because terrestrial backgrounds are considerably more complex than the relatively smooth background of space. Additionally, only a few artificial stars may be added to the image data at any one time, otherwise the artificial stars themselves significantly change the crowding of the field and the results become unreliable. Thus, a full data set (plus artificial stars) must be analyzed many, potentially thousands of times using, for example, Monte Carlo simulation, to obtain a large statistical sample of artificial stars. Conceptually, the artificial star population must be comparable to the population of real stars, otherwise biases will be introduced. However, the population of real stars is not known in advance, so ensuring that the artificial stars are truly comparable to real stars requires considerable ingenuity. Additionally, there are a variety of problems of detail, such as the best way to characterize the light distribution that would be created by an individual star. Moreover, determining the validity of each round of artificial star analysis is labor intensive and time consuming. Historically, each new round of artificial star analysis was preceded by an astronomer evaluating the previously acquired data set to determine if the resulting implied confidence-level was achieved. The astronomer would then decide whether additional Monte Carlo simulations were necessary.
Benefits of artificial star analysis include determination of detection probabilities across a range of observing conditions (light-level, background/foreground image structure, high-noise, etc.).